Data processing apparatus

Abstract

Data processing apparatus and methods are provided. One data processing apparatus comprises: a plurality of pipelined stages, each of the plurality pipelined stages being operable in each processing cycle to receive a group of data elements from an earlier pipelined stage; permute logic operable to buffer ‘n’ of the groups of data elements over a corresponding ‘n’ processing cycles thereby creating a bubble within pipelined stages, and forwarding logic operable, once the ‘n’ of the groups of data elements have been buffered by the permute logic, to forward permuted groups of data elements comprising the data elements reordered by the permute logic to fill the bubble within the pipelined stages. By forwarding the data elements to fill the bubble an improved throughput can be achieved and since a constant stream of data can be transformed without the need to increase the number of input or output registers required to support the permute logic, the need to duplicate the permute logic or the need to introduce any additional storage elements.

Description

BACKGROUND OF THE INVENTION [0001] 1. Field of the Invention [0002] The present invention relates to a data processing apparatus and method. Embodiments of the present invention relate to a data processing apparatus and method operable to perform permute operations. [0003] 2. Description of the Prior Art [0004] Permute operations are known. Permute operations typically take a sequence of data elements and reorder or permutate the data elements to create a new sequence. [0005] For example, as shown in FIG. 1, a sequence of consecutive data elements 0 to 15 are provided. A permute unit 10 is provided which performs a permute operation on the data elements in response to a permute instruction. Such an instruction is typically supported by a vector or single instruction multiple data (SIMD) data processing apparatus for supporting transformation between arrays of structures (AoS) and structures of arrays (AoS). UK patent application 2,409,063 filed on 9 Dec. 2003 by ARM Limited describe...

AI Technical Summary

Benefits of technology

[0018] The present invention also recognises that whilst it would be possible to maximise throughput by, for example, increasing the number of registers which feed the permute logic, increasing the number of registers which receive the permuted data elements from the permute logic or by duplicating the permute logic or the registers used to buffer the data elements to be permuted, such an approach undesirably increases the amount of resources required.

[0041] Accordingly, the permute logic takes advantage of the fact that the size of the bubble created within the pipeline stages will match the number of permuted data elements required to be provided from the permute logic, thereby filling the bubble.

Problems solved by technology

The present invention also recognises that whilst it would be possible to maximise throughput by, for example, increasing the number of registers which feed the permute logic, increasing the number of registers which receive the permuted data elements from the permute logic or by duplicating the permute logic or the registers used to buffer the data elements to be permuted, such an approach undesirably increases the amount of resources required.

Images